Research on audio characteristics based on auditory characteristics and instantaneous frequency

This paper proposes a novel audio feature extraction method that integrates auditory characteristics and instantaneous frequency, aiming to enhance the representation of audio signals. While traditional Mel-scale Frequency Cepstral Coefficient features partially capture human auditory perception, their filter shape significantly deviates from the actual auditory filters. To address this limitation, we introduce the GammaTone filter in conjunction with equivalent rectangular bandwidth scale and instantaneous frequency, resulting in two new feature sets: GammaTone Frequency Cepstral Coefficients and Instantaneous Frequency GammaTone Frequency Cepstral Coefficients. Experimental results demonstrate that both GammaTone Frequency Cepstral Coefficients and Instantaneous Frequency GammaTone Frequency Cepstral Coefficients exhibit superior capabilities in representing non-speech audio signals.